Qualcomm AI Engine Direct - support dim_order

[haowhsu-quic]

Summary

• remove _to_dim op and let backend take care of tensor layout
  • Current layout_transform works well.
  • We always use contiguous memory to fetch parameters.
  • Remove this memory format operator won't affect the original output.
• test cases

Fixes #<4873>

Test plan

python backends/qualcomm/tests/test_qnn_delegate.py -k TestQNNQuantizedOperators.test_qnn_backend_conv2d_channel_last -s $DEVICE_SERIAL -m $SOC_MODEL -b build-android/ -a $PATH_TO_ARTIFACTS

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/7189

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 1 New Failure

As of commit 6f54df9 with merge base 8861b9a ():

NEW FAILURE - The following job has failed:

• pull / unittest-arm / linux-job (gh)
  RuntimeError: Command docker exec -t 01c02038e819a0c5d8cc8c6b785e054635c349a6e2c426d5afc85740d9284226 /exec failed with exit code 1

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[facebook-github-bot]

@cccclai has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.